Read and parse

What it does

An AI OCR model reads every page of an uploaded file and turns it into text that fileAI classifies and extracts from. You pick the model per upload, or rely on your project default, and read the page text back with GET /v1/files/{fileId}/ocr.

When to use it

You receive documents in Japanese, Chinese or Thai and need a model built for that language instead of the English default.
You need the raw text of each page, for example to index it for search or run your own parsing next to fileAI’s extraction.
You want every new file in a project read by a different model, without changing each upload request.

How it works

fileAI picks the model

If the upload request includes ocrModel, fileAI uses that model for the file. Otherwise it uses your project default.

The model reads each page

The text it finds is what fileAI classifies and extracts from.

You read the text

GET /v1/files/{fileId}/ocr returns the text of each page, using a fileId from the upload’s files[].

Try it

Read the page text of a processed file:

curl "$FILEAI_BASE_URL/files/<fileId>/ocr" \
  -H "x-api-key: $FILEAI_API_KEY"

import os, requests

BASE = "https://api.orion.file.ai/prod/v1"
HEADERS = {"x-api-key": os.environ["FILEAI_API_KEY"]}

ocr = requests.get(f"{BASE}/files/{file_id}/ocr", headers=HEADERS).json()
for page in ocr["ocr"]:
    print(page["page_number"], page["text"][:80])

const BASE = "https://api.orion.file.ai/prod/v1";
const headers = { "x-api-key": process.env.FILEAI_API_KEY };

const ocr = await fetch(`${BASE}/files/${fileId}/ocr`, { headers }).then((r) => r.json());
for (const page of ocr.ocr) console.log(page.page_number, page.text.slice(0, 80));

Understand the result

Response (200)

{
  "ocr": [
    { "page_number": 1, "text": "OCR Result page 1 example" },
    { "page_number": 2, "text": "OCR Result page 2 example" }
  ]
}

Field	Meaning
`ocr[]`	One entry per page.
`ocr[].page_number`	The page the text came from.
`ocr[].text`	The page’s text as plain text.

The response is plain text per page only. It has no layout, bounding boxes or coordinates.

A fileId that is blank or not found in your project returns 422 with Invalid fileId.

Next steps

Uploads

Where to send ocrModel and the other upload fields.

Supported file types

The files each model can read.

Extract with AI Schemas

Turn the text into structured fields.

Reference

How the model is chosen

If the upload request includes ocrModel, fileAI uses that model for the file.
Otherwise fileAI uses your project default. Unless your project is configured otherwise, that is Beethoven_ENG_GP25.2. To change it, see File intelligence.

ocrModel is accepted on every upload route: POST /v1/files/upload, /files/upload/batch (per file or in defaults), /files/upload/multipart and /files/upload/multipart/batch. Send the model name exactly as listed below.

Available models

Model name	Use it for
`Beethoven_ENG_O5.6`	English documents (OpenAI v6).
`Beethoven_ENG_G5.5`	English documents (Gemini v5).
`Beethoven_ENG_GP25`	English documents (Gemini Pro 2.5).
`Beethoven_ENG_GP25.1`	English documents (Gemini Pro 2.5 v1).
`Beethoven_ENG_GP25.2`	English PDFs (Gemini Pro 2.5 PDF). The default unless your project is configured otherwise.
`Unified (google-document-ai-ocr-gemini-v10)`	Unified model.
`Aegis (google-document-ai-ocr-gemini-aegis-v1)`	Aegis model.
`Beethoven_Direct_Form_Filling (GP2.5)`	Gemini Pro 2.5 OCR with direct form filling turned on for your project.
`Beethoven_CUS_O5.1`	Custom model (OpenAI v8).
`Beethoven_CUS_O5.2`	Custom model (Gemini v13).
`Beethoven_CUS_GP25.1`	Custom model (Gemini Pro 2.5).
`Beethoven_ZH_O5.9`	Chinese documents (OpenAI v9).
`Beethoven_JP_O5.3`	Japanese documents (OpenAI v3).
`Beethoven_JP_G5.4`	Japanese documents (Gemini, fine-tuned).
`Beethoven_TH_O5.1`	Thai documents (OpenAI v1).
`Beethoven_TH_G5.1`	Thai documents (Gemini v1).
`Capybara (Experimental)`	Experimental model. Also accepted as `capybara`. Only for projects set up for it. See Experimental model.

Experimental model

Capybara (Experimental) is available only to projects set up for it, in organizations where it is enabled. A per-upload ocrModel must match the project’s setup, or the request returns 422:

In a project not set up for it, capybara is rejected.
In a project set up for it, any other model is rejected. Omit ocrModel to use the project setting.

Retired models

Models that are no longer offered return 422 when you send them, on uploads and on PATCH /v1/project/setting:

{ "status": 422, "detail": "OCR model 'Beethoven_ENG_GP3' has been retired." }

Retired names include Beethoven_ENG_GP3, Beethoven_ENG_O5.1 to O5.5, Beethoven_ENG_G5.0 and G5.2 to G5.4, Beethoven_TH_O5.0, Beethoven_JP_O5.1, Beethoven_JP_O5.2 and Beethoven_JP_G5.1 to G5.3. A name that was never a model returns 422 with Invalid ocr model.

Override the model for one upload

curl -X POST "$FILEAI_BASE_URL/files/upload" \
  -H "x-api-key: $FILEAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "fileName": "invoice-jp.pdf",
    "fileType": "application/pdf",
    "ocrModel": "Beethoven_JP_G5.4"
  }'

Change the project default

The project default model is set in the app or with PATCH /v1/project/setting. See File intelligence. New defaults apply to files processed after the change.

Get started

Products

Build

Go live

What it does

When to use it

How it works

Try it

Understand the result

Next steps

Uploads

Supported file types

Extract with AI Schemas

Reference

How the model is chosen

Available models

Experimental model

Retired models

Override the model for one upload

Change the project default

​What it does

​When to use it

​How it works

​Try it

​Understand the result

​Next steps

Uploads

Supported file types

Extract with AI Schemas

​Reference

​How the model is chosen

​Available models

​Experimental model

​Retired models

​Override the model for one upload

​Change the project default

What it does

When to use it

How it works

Try it

Understand the result

Next steps

Reference

How the model is chosen

Available models

Experimental model

Retired models

Override the model for one upload

Change the project default